In this episode of the Mostly Unstructured podcast, Ed and Clay discuss whether it’s better to train a domain‑specific LLM or leverage foundational models like ChatGPT, Gemini and Claude. They explain the trade‑offs between fine‑tuning and retrieval‑augmented generation (RAG), and why Intelligent Document Processing (IDP) is vital for turning unstructured data into usable context. In this discussion, we cover:
- Why training your own LLM is risky and often unnecessary compared to adopting and building from a foundational model.
- How retrieval‑augmented generation (RAG) delivers more accurate results than simple fine‑tuning.
- The importance of Intelligent Document Processing (IDP) for ingesting unstructured data and building domain context.
- Real‑world lessons on AI governance, including the Air Canada bereavement‑policy chatbot case.
- Managing bias, hallucinations and toxicity in enterprise models.
- Measuring your return on AI investment.
